beamrider_ppo.yaml

alg_para:
  alg_name: PPO

env_para:
  env_name: AtariEnv
  env_info:
    name: BeamRiderNoFrameskip-v4
    vision: False

agent_para:
  agent_name: AtariPpo
  agent_num : 1
  agent_config:
    max_steps: 128
    complete_step: 10000000

model_para:
  actor:
    model_name: PpoCnn
    state_dim: [84, 84, 4]
    action_dim: 9
    input_dtype: uint8
    model_config:
      BATCH_SIZE: 320
      CRITIC_LOSS_COEF: 1.0
      ENTROPY_LOSS: 0.003
      LOSS_CLIPPING: 0.1
      LR: 0.00025
      MAX_GRAD_NORM: 5.0
      NUM_SGD_ITER: 4
      SUMMARY: False
      VF_SHARE_LAYERS: True
      activation: relu
      hidden_sizes: [512]

env_num: 10

benchmark:
  log_interval_to_train: 10